CLASSIFICATION AND RULE EXTRACTION USING ARTIFICIAL NEURAL NETWORKS
Loading...
Date
2001-12
Authors
TAN, SHING CHIANG
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This thesis is concerned with the development of neural-network-based autonomous
learning systems for pattern classification and rule extraction. Two aspects of neural
networks are given prime attention, viz. their incremental learning ability and their
capability of presenting knowledge in the form of fuzzy rules. A novel rule-based
network, which incorporates both the above aspects, is devised using existing
methodologies. In addition, a real-world case study in collaboration with a power
generation station, is conducted. The system under investigation is the Circulating
Water (CW) system. In this case, the rule-based network is employed to handle fault
detection and diagnosis problems of the CW system.
The incrementally learning ability is adopted from the family of Adaptive
Resonance Theory (ART) networks. In particular, the Fuzzy ARTMAP (FAM) network
is used as a platform for the rule-based system developed here. The ability of F AM in
classification tasks is first examined. Following this, the rule extraction algorithm of
F AM that extracts the knowledge embedded in the network is applied. Nevertheless,
the extracted rule set exhibits several limitations. As a result, FAM-RecBFN, a hybrid
network, based on F AM and the Rectangular Basis Function Network, is introduced.
To demonstrate the effectiveness of the proposed FAM-RecBFN, a number of
simulations are conducted usmg benchmark problems. Then the F AM-RecBFN is
applied to fault classification and diagnosis of the CW system. Several data processing
strategies, such as randomised and ordered input data presentation to the F AMRecBFN,
are investigated. The experimental results show the potential of the F AMRecBFN
system as a fault diagnosis tool with an explanatory facility for safety critical
applications, such as in power generation plants.
Description
Keywords
CLASSIFICATION , RULE EXTRACTION